1

RH: Cognition Everywhere

Cognition Everywhere:
The Rise of the Cognitive Nonconscious and the Costs of Consciousness
N. Katherine Hayles

A massive shift is underway in our intellectual and cultural formations. Many
different streams of thought are contributing, coming from diverse intellectual traditions,
holding various kinds of commitments, and employing divergent methodologies. The
differences notwithstanding, they agree on a central tenet: the importance of
nonconscious cognition, its pervasiveness and computational potential, and its ability to
pose new kinds of challenges not just to rationality but to consciousness in general,
including the experience of selfhood, the power of reason, and the evolutionary costs and
systemic blindnesses of consciousness.
The implications for interpretation are profound. Interpretation is deeply linked
with questions of meaning; indeed, many dictionaries define interpretation in terms of
meaning (and meaning in terms of interpretation). For the cognitive nonconscious,
however, meaning has no meaning. As the cognitive nonconscious reaches
unprecedented importance in communication technologies, ambient systems, embedded
devices, and other technological affordances, interpretation has become deeply entwined
with the cognitive nonconscious, opening new avenues for exploring, assessing, debating
and resisting possible configurations between interpretive strategies and the cognitive
nonconscious. In a later section I will identify sites within the humanities where these
2
contestations and reconfigurations are most active and comment upon the strategies
emerging there. Whatever one makes of these changes, one conclusion seems
inescapable: the humanities cannot continue to take the quest for meaning as an
unquestioned premise for their ways of doing business. Before we arrive at this point,
some ground clearing of terminology is necessary, as well as consideration of how the
cognitive nonconscious differs from and interacts with consciousness.

Rethinking the Cognitive Nonconscious
One way into understanding the cognitive nonconscious is through Stanis!aw
Lems Summa Technologiae,
1
a work that, as far as the Anglophone world is concerned,
has been caught in a time-warp. Published in Polish in 1964, Summa was never
completely translated into English prior to its present appearance in 2013. Lem
presciently understood that our society was facing what he called an information
barrier, a deluge of information that would overwhelm scientific and technological
enterprises unless a way was found to automate cognition. He observed that formal
languages, such as mathematics and explicit equations, do not deal well with complexity.
For example, equations for gravitational interactions do not have explicit solutions when
as few as three bodies are involved. Nevertheless, there are many instances when
complex problems are solved effortlessly by nonconscious means. For example, when a
rabbit chased by a coyote leaps over a chasm, the feat would require many equations and
considerable time to solve explicitly, but the animal does it instantly without a single
calculation.
S
Reasoning that translating tasks into formal languages may be unnecessary for
solving complex problems, Lem proposes a form of evolutionary computation
programmed in natural media, an information farm in which systems could
successfully perform cognitive modeling functions without consciousness. He suggests
modeling a dynamic system by first creating a diversity generator such as a fast-
running stream carrying along rocks of various sizes. Then, to match the target systems
momentum, one places some kind of barrier or sieve that selects only rocks of a
specific size and velocity. Other sieves select for different variables, and the process
continues until the desired match is achieved.
2
One could imagine expanding this kind
of modeling by using living cells, which in carrying out division, excretion, and osmosis
employ many different kinds of sieves as selection devices. In fact, contemporary
experiments by Leonard Adleman show it is possible to use DNA sequences to solve
complex topological challenges similar to the traveling salesman problem, another
example of how the cognitive nonconscious can be harnessed to arrive at solutions
difficult or impossible to achieve by explicit means.
What kind of processes do such systems entail, and what is implied by calling
them cognitive? First, these systems operate within evolutionary dynamics, that is, they
are subjected to fitness criteria that select certain states out of the diverse range available.
Second, they are adaptive; they change their behaviors as a result of fitness challenges
such as homeostasis for the cell. Third, they are complex, composed of parts interacting
with each other in multiple recursive feedback loops, or what Andy Clark calls
continuous reciprocal causation.
3
Consequently, they exhibit emergence, results that
cannot be predicted and that exceed the sum of their parts. Fourth, they are constraint
4
driven, which implies that the individual agents behaviors are guided by simple
instincts or rules that constrain them to certain productive paths, such as the sieves
mentioned above. Generally, they enact the artificial life mantra: From simple rules to
complex patterns or behaviors. Together, these properties enable such systems to
perform modeling and other functions that, if they were performed by a conscious entity,
would unquestionably be called cognitive.
To avoid confusion, I will reserve thinking for what conscious entities such as
humans (and some animals) do, and cognition as a broader term that does not
necessarily require consciousness but has the effect of performing complex modeling and
other informational tasks. On this view, we can say that while all thinking is cognition,
not all cognition is thinking. In this respect the cognitive nonconscious is qualitatively
different from the unconscious, which communicates with consciousness in a number of
ways.
4
Accordingly, I will call consciousness/unconsciousness modes of awareness.
By contrast, the cognitive nonconscious operates at a lower level of neuronal
organization not accessible to introspection.
Nonconscious cognitive systems are distinct from the processes that generate
them because they show an intention toward not present in the underlying material
processes as such. For example, a termite mound is a complex architectural structure that
emerges as a result of pheromone trails laid down by individual termites enacting simple
behavioral rules.
5
It has an intention toward, namely the protection and preservation of
the colony. Another example is a beehive, an emergent result created when individual
bees position themselves at a certain distance from their fellows and, moving in a circle,
spit wax. Since the adjacent bees are doing the same, the wax lines press against each
S
other and form a hexagon, the polygon with the most efficient packing.
6
The intention
toward is instantiated in the beehive, the emergent result of individual bees acting as
autonomous individual agents, each of which need only perform a few simple behaviors
to achieve an effect greater than the sum of the beehives parts.
In contrast to the cognitive nonconscious, material processes operating on their
own rather than as part of a complex adaptive system do not demonstrate emergence,
adaptation, or complexity. For example, a glacier sliding downhill generally lacks
adaptive behavior (it cannot choose a shady versus a sunny valley), has negligible
emergent capacity, and its path can be calculated precisely if the relevant forces are
known. The distinction between material processes and complex systems may not
always be so clear cut. Indeed, this framework, positing a tripartite structure of conscious
thinking, nonconscious cognition, and material processes, catalyzes boundary questions
about the delineations between categories as active sites for interpretation and debate.
Enlarged beyond its traditional identification with thought, cognition in some
instances may be located in the system rather than an individual participant, an important
change from a model of cognition centered in the self. As a general concept, the term
cognitive nonconscious does not specify whether the cognition occurs inside the mental
world of the participant, between participants, or within the system as a whole. It may
operate wholly independently from consciousness, as in the cases of bees and termites, or
it may be part of the larger system such as a human, where it mediates between material
processes and the emergence of consciousness/unconsciousness. Alternatively, it may be
instantiated in a technological device such as a computer. Nonconscious cognition, then,
6
operates across and within the full spectrum of cognitive agents: humans, animals, and
technical devices.

The Costs of Consciousness
Along with an expanded sense of cognition come reassessments of consciousness,
the purposes it serves, and the costs it entails. Most researchers recognize (at least) two
levels of consciousness, a lower level called core or primary consciousness, and a higher
level called extended or secondary consciousness. Humans share core consciousness
with other primates and (arguably) a wide range of mammals and other animals as well.
According to Thomas Metzinger
7
, a contemporary German philosopher, core
consciousness creates a mental model of itself that he calls a phenomenal self-model
(PSM) (107); it also creates a model of its relations to others, the phenomenal model of
the intentionality relation (PMIR) (301-5). Neither of these models could exist without
consciousness, since they require the memory of past events and the anticipation of future
ones. From these models, the experience of a self arises, the feeling of an I that persists
through time and has a more or less continuous identity. The PMIR allows the self to
operate contextually with others with whom it constructs an intentionality relation.
The sense of self, Metzinger argues, is an illusion, facilitated by the fact that the
construction of the PSM and the PMIR models are transparent to the self (that is, the self
does not recognize them as models but takes them as actually existing entities). This
leads Metzinger to conclude, nobody ever was or had a self (1). In effect, by
positioning the self as epiphenomenal, he reduces the phenomenal experience of self back
to the underlying material processes. Philosopher of consciousness Owen Flanagan,
7
following William James, tracks a similar line of reasoning: the self is a construct, a
model, a product of conceiving of our organically connected mental life in a certain
way.
8
Who thinks the thoughts that we associate with the self? According to Flanagan
(and James), the thoughts think themselves, each carrying along with it the memories,
feelings, and conclusions of its predecessor while bearing them toward its successor.
Antonio Damasio holds a somewhat similar view, in the sense that he considers
the self to be a construct created through experiences, emotions, and feelings a child has
as she grows rather than an essential attribute or possession. Damasio, however, also
thinks that the self (illusion though it may be) evolved because it has a functional
purpose, namely to create a concern for preservation and well-being that propels the
organism into action and thus guarantees that proper attention is paid to the matters of
individual life.
9
Owen Flanagan agrees: consciousness and the sense of self have
functions, including serving as a clearinghouse of sorts where past experiences are
recalled as memories and future anticipations are generated and compared with memories
in order to arrive at projections and outcomes. In Daniel Dennetts metaphor,
consciousness and the working memory it enables constitute the workspace where past,
present, and future are put together to form meaningful sequences.
10

Meaning, then, can be understood at the level of core consciousness as an
emergent result of the relation between the PSM and the PMIRthat is, between the self-
model and the models the self constructs of objects which it has an intention toward.
Damasio puts it more strongly; there is no self without awareness of and engagement
with others.
11
The self thus requires core consciousness, which constructs the PSM and
the PMIR; without consciousness, a self could not exist. In humans (and some animals),
8
the core self is overlaid with a higher-level consciousness capable of metalevel reasoning,
including interrogations of meanings that call for interpretations.
In addition to concern for the self, a crucial role of consciousness, which occurs at
both the core and metalevel, is creating and maintaining a coherent picture of the world.
As Gerald Edelman and Giulio Torino put it, Many neuropsychological disorders
demonstrate that consciousness can bend or shrink, and at times even split but it does not
tolerate breaks of coherence.
12
We can easily see how this quality would have adaptive
advantages. Creating coherence enables the self to model causal interactions reliably,
make reasonable anticipations, and smooth over the gaps and breaks that phenomenal
experiences present. If a car is momentarily hidden by a truck and then reappears,
consciousness recognizes this as the same car, often at a level below focused attention.
This very quality, however, also frequently causes consciousness to misrepresent
anomalous or strange situations.
A number of experiments in cognitive psychology confirm this fact. In one now-
famous situation,
13
subjects are shown a video of players passing a basketball and are
asked to keep track of the passes. In the middle of the scene, someone dressed in a
gorilla suit walked across the playing area, but a majority of subjects report that they saw
nothing unusual.
14
In another staged situation, a man stops a passerby and asks for
directions.
15
While the subject is speaking, two workmen carrying a vertical sheet of
wood pass between them, momentarily blocking the view. When they pass, the
interlocutor has been replaced by another person, but the majority of subjects do not
notice the discrepancy. Useful as is the tendency of consciousness to insist on coherence,
these experiments show that one cost is the screening out of highly unusual events.
9
Without our being aware of it, consciousness edits to make them conform to customary
expectations, a function that makes eyewitness testimony notoriously unreliable. Even in
the most ordinary circumstances, consciousness confabulates more or less continuously,
smoothing out the world to fit our expectations and screening from us the worlds
capacity for infinite surprise.
A second cost is the fact that consciousness is slow relative to perception.
Experiments by Benjamin Libet and colleagues show that before subjects indicate that
they have decided to raise their arms, the muscle action has already started.
16
Although
Daniel Dennett is critical of Libets experimental design, he agrees that consciousness is
belated, behind perception by several hundred milliseconds, the so-called missing half-
second.
17
This cost, although negligible in many contexts, assumes new importance
when cognitive nonconscious technical devices can operate at temporal regimes
inaccessible to humans and exploit the missing half second to their advantage.
Finally there are the costs, difficult to calculate, of possessing a self aware of
itself and tending to make that self the primary actor in every scene. Damasio comments
that consciousness, as currently designed, constrains the world of imagination to be first
and foremost about the individual, about an individual organism, about the self in the
broad sense of the term.
18
The anthropocentric bias for which humans are notorious
would not be possible, at least in the same sense, without consciousness and the
impression of a reified self that consciousness creates. The same faculty that makes us
aware of ourselves as selves also partially blinds us to the complexity of the biological,
social, and technological systems in which we are embedded, tending to make us think
we are the most important actors and that we can control the consequences of our actions
1u
and those of other agents. As we are discovering, from climate change to ocean
acidification to greenhouse effects, this is far from the case.

Neural Correlates to Consciousness and the Cognitive Nonconscious
Damasio and Edelman, two eminent neurobiologists, have complementary research
projects, Damasio working from brain macrostructures on down, Edelman working from
brain neurons on up. Together, their research presents a compelling picture of how core
consciousness connects with the cognitive nonconscious. Damasios work has been
especially influential in deciphering how body states are represented in human and
primate brains through somatic markers, indicators emerging from chemical
concentrations in the blood and electrical signals in neuronal formations.
19
In a sense,
this is an easier problem to solve than how the brain interacts with the outside world,
because body states normally fluctuate within a narrow range of parameters consistent
with life; if these are exceeded, the organism risks illness or death. The markers, sending
information to centers in the brain, help initiate events such as emotionsbodily states
corresponding to what the markers indicateand feelings, mental experiences that signal
such sensations as feeling hungry, tired, thirsty and frightened.
From the parts of the brain registering these markers emerge what Damasio calls the
protoself, an interconnected and temporarily coherent collection of neural patterns
which represent the state of the organism, moment by moment, at multiple levels of the
brain (174). The protoself, Damasio emphasizes, instantiates being but not
consciousness or knowledge; it corresponds to what I have been calling the cognitive
nonconscious. Its actions may properly be called cognitive in my sense because it has an
11
intention toward, namely the representation of body states. Moreover, it is embedded
in highly complex systems that are both adaptive and recursive. When the organism
encounters an object, which Damasio refers to as something-to-be-known, the object
is also mapped within the brain, in the sensory and motor structures activated by the
interaction of the organism with the object (169). This in turn causes modifications in
the maps pertaining to the organism and generates core consciousness, a recursive cycle
that can also map the maps in second-order interactions and thereby give rise to extended
consciousness. Consciousness in any form only arises, he maintains, when the object,
the organism, and their relation, can be re-represented (160). Obviously, to be re-
represented, they must first have been represented, and this mapping gives rise to and
occurs within the protoself. The protoself, then, is the level at which somatic markers are
assembled into body maps, thus mediating between consciousness and the underlying
material processes of neuronal and chemical signals.
This picture of how consciousness arises finds support in the work of Nobel Prize
winner neurologist Gerald M. Edelman and his colleague Giulio Tononi.
20
Their
analysis suggests that a group of neurons can contribute to the contents of consciousness
if and only if it forms a distributed functional cluster of neurons interconnected within
themselves and with the thalamocortical system, achieving a high degree of interaction
within hundreds of milliseconds. Moreover, the neurons within the cluster must be
highly differentiated, leading to high values of complexity (146).
To provide a context for these conclusions, we may briefly review Edelmans
theory of neuronal group selection (TNGS), which he calls neural Darwinism.
21
The
basic idea is that functional clusters of neurons flourish and grow if they deal effectively
12
with relevant sensory inputs; those less efficient tend to dwindle and die out. In addition
to the neural clusters, Edelman (like Damasio) proposes that the brain develops maps, for
example, clusters of neurons that map input from the retina. Neural groups are connected
between themselves through recursive reentrant connections (45-50, esp. 45), flows of
information from one cluster to another and back through massively parallel connections.
The maps are interconnected by similar flows, and maps and clusters are also connected
to each other.
To assess the degree of complexity that a functional neuronal cluster possesses,
Edelman and Tononi have developed a tool they call the functional cluster index (CI).
22

This concept allows a precise measure of the relative strength of causal interactions
within elements of the cluster compared to their interactions with other neurons active in
the brain. A value of CI = 1 means that the neurons in the cluster are as active with other
neurons outside the cluster as they are among themselves. Functional clusters
contributing to consciousness have values much greater than one, indicating that they are
strongly interacting among themselves and only weakly interacting with other neurons
active at that time.
From the chaotic storm of firing neurons, the coherence of the clusters mobilize
neurons from different parts of the brain to create coherent maps of body states, and these
maps coalesce into what Edelman calls scenes, which in turn coalesce to create what he
calls primary consciousness (in Damasios terms, core consciousness). Edelmans
account adds to Damasios the neuronal mechanisms and dynamics that constitute a
protoself from the underlying neurons and neuronal clusters, as well as the processes by
1S
which scenes are built from maps through recursive interactions between an organisms
representations of body states and representations of its relations with objects.
It is worth emphasizing the central role that continuous reciprocal causation plays in both
Damasios and Edelmans accounts. Thirty years ago, Humberto Maturana and
Francisco Varela intuited that recursion was central to cognition,
23
a hypothesis now
tested and extended through much-improved imaging technologies, microelectrode
studies, and other contemporary research practices.
Let us now turn to the processes by which re-representation occurs. Recalling
Damasios strong claim that there is no consciousness without re-representation, re-
representation is clearly a major function of the protoself, site of the cognitive
nonconscious and the processes that give rise to core and higher consciousness. In his
theory of grounded cognition, Lawrence W. Barsalou in an influential article gives a
compelling account of how re-representation occurs in what he calls simulation, the
re-enactment of perceptual, motor, and introspective states acquired during experience
with the world, body, and mind.
24

In particular, sensory experiences are simulated when concepts relevant to those
experiences are processed and understood. He marshals a host of experimental evidence
indicating that such mental re-enactments are integral parts of cognitive processing,
including even thoughts pertaining to highly abstract concepts. The theory of grounded
cognition reflects the assumption that cognition is typically grounded in multiple ways,
including simulations, situated action, and, on occasion, bodily states (619). For
example, perceiving a cup handle triggers simulations of both grasping and functional
actions, as indicated by fMRI scans (functional magnetic resonance images). The
14
simulation mechanism is also activated when the subject sees someone else perform an
action; accurately judging the weight of an object lifted by another agent requires
simulating the lifting action in ones own motor and somatosensory systems (624). In
order for a pianist to identify auditory recordings of his own playing, he must simulate
the motor actions underlying it (624). Perhaps most surprising, such simulations are
also necessary to grasp abstract concepts, indicating that thinking is deeply entwined with
the recall and reenactment of bodily states and actions. The importance of simulations in
higher-level thinking shows that biological systems have evolved mechanisms to re-
represent perceptual and bodily states, not only to make them accessible to consciousness
but also to support and ground thoughts related to them. These thoughts in turn feed back
to affect somatic states. We can now appreciate the emphasis that Damasio places on re-
representation, for it serves as an essential part of the communication processes between
the protoself and consciousness, and it also invests abstract thought with grounding in
somatic states.
Although the reenactment mechanisms differ with present-day computational
methods, the idea of recursion is central in artificial media as well. Whereas with
biological organisms bodily states provide the basis for higher-level thinking, with
artificial media recursion operates along a hierarchy that moves from simple to complex,
local individual agents operating according to a few simple rules to global systemic
patterns of complexity. For technical devices, an intention toward is necessary but not
sufficient: a hammer and a finance trading algorithm are both designed with an intention
in mind, but only the trading algorithm demonstrates nonconscious cognition. What
makes the difference? Nonconscious cognition operates through many of the same
1S
strategies employed by biological organisms, including emergence (call it the termite
strategy), re-representation (as in grounded cognition), evolutionary dynamics to
bootstrap cognition, and a variety of other mechanisms. Some nonconscious cognitive
devices have sensors and actuators, so they can interact with their environments and
perform actions in the world. Others live in artificial environments structured to judge
performances according to predetermined fitness criteria, allowing only the most
successful agents to propagate into the next generation. Although the range of technical
devices demonstrating nonconscious cognition is too broad to cover here, the next section
will give a sense of their range and diversity.

Technical Devices and the Cognitive Nonconscious
Evolutionary computations, so called because they instantiate evolutionary
algorithms in a variety of artificial media, have by now been extensively studied. John R.
Koza and coauthors have created genetic algorithms to carry out a variety of tasks, for
example designing electric circuits.
25
Their work demonstrates typical strategies to
achieve nonconscious cognition. The seed program generates an array of very simple
circuits. The performance of each circuit is tested according to how well it carries out
certain tasks. The most successful are selected and married to each other (that is, their
circuits are combined to create hybrids) which are used to create the next generation,
circuits somewhat similar to the parents but with minor variations among the children.
The most successful of these are again selected and again propagate with minor
variations, and so on through hundreds or thousands of generations. Eventually circuits
evolve that can achieve what Koza and his colleagues call human-competitive results
16
(1), which they define as designs publishable in a peer-reviewed professional journal or
circuits judged worthy of a patent.
Similar techniques have been used with algorithms designed to compose music.
Such programs are typically given predefined grammars, but they also can modify these
or even create their own grammar. As noted by John A. Maurer in A Brief History of
Algorithmic Composition,
26
a genetic algorithm system created by David Cope, called
Experiments in Musical Intelligence (EMI), works with a large database of different
composition strategies, which it can draw on and/or modify. Creating from scores fed
into it, it can also create its own database and compose based on that. Compositions in
the style of a number of composers have been created in this fashion, including Bach,
Mozart, Brahms, and others. There are also genetic programs that start with a small
number of functions such as transposition, note generation, and creating or modifying
time values. The program then randomly combines these functions, which are judged
according to some fitness value. The most successful are married, as with Kozas
genetic algorithms, and produce children which are evaluated against fitness values in
turn to identify the next pair of parents, and so forth.
Recently programs have been developed that act as critics judging how
commercially successful a given composition (or movie) is likely to be. Christopher
Steiner discusses the case of Polyphonic HMI,
27
a company that developed algorithms to
evaluate the likely commercial success of a song. The algorithm works by using Fourier
transforms and other mathematical functions to isolate and analyze tempos, melodies,
beats, rhythms and so forth, creating a three-dimensional visualization showing how
similar the song is to songs that have made it big in the past. Mike McCready, creator of
17
the Polyphonic HMI, used the program to assess an album by Norah Jones, then an
unknown artist, and discovered it had extraordinarily large fitness values. Subsequently,
the album went on to sell twenty million copies and won eight Grammy Awards (83).
Other kinds of evolving agents employ learning processes similar to embodied
biological organisms, using their experiences in the world as physical beings to learn,
draw inferences, achieve simple linguistic skills, and interact with humans. Rodney
Brooks Cog, a head and torso robot, exemplifies this kind of approach (begun in 1994,
Cog was retired in 2003). Brooks advocates what he calls cheap tricks, emergent results
caused by the interactions of different systems within the robot, often giving the
appearance of human-level intelligence without, however, possessing any conscious
awareness.
28
Another version of a language-learning device is Tom Mitchells NELL
(Never-Ending Language Learning), a program that scans wild (i.e., unstructured) text
on the internet 24/7 and draws inferences from it with a minimum of human supervision
(http://www.cmu.edu/homepage/computing/2010/fall/nell-computer-that-learns.shtml).
29

Less exotic everyday software demonstrating some of the same properties are programs
that draw inferences from databanks about a users preferences, for example programs
used by Amazon to make suggestions for future purchases (We think you might
like . . .).
In the financial markets, automated trading algorithms, which now account for
about 70 percent of all trades, also operate in highly competitive ecologies.
30
The faster
algorithms can detect pending orders from their slower competitors and front run their
orders, for example by purchasing a desired stock at a lower price and then, within
milliseconds, turning around and offering it at a slightly higher price, which the slower
18
algorithm now has no choice but to buy at the new price. Such algorithms typically have
several trading strategies from which to choose, and they will opt for the one that yields
the best final result. In addition, capitalizing on market regulations, they also use their
temporal advantages to force other algorithms to be charged fees while raking in rebates
for themselves (the so-called maker and taker fees and rebates).
The example of trading algorithms demonstrates that, when nonconscious
cognitive devices penetrate far enough into human systems, they can potentially change
the dynamics of human behaviors. As Neil Johnson and his collaborators at Nanex argue
(a firm specializing in studying the behaviors of automated trading algorithms), the effect
of automated trading algorithms has transformed the stock market from a mixed human-
machine ecology to a machine-machine ecology.
31
As algorithms account for more and
more trades, the major exchanges (now for-profit corporations themselves) shape their
practices accordingly, for example by offering to sell at a premium rack space next to
their servers, thereby shaving milliseconds off the transmission time, a temporal interval
in which further financial advantages can be gained. The exchanges have also multiplied
the kinds of bids that can be submitted, giving algorithms more ways to turn milliseconds
into megadollars.
The ways in which built environments affect human cognition have of course
been extensively studied in architecture, geography, economics, political science, group
psychology, and a host of other fields. It is scarcely news that humans are affected not
only by social exchanges with each other but also by their interactions with their
environments. What is (relatively) new is the extent to which the built environment
instantiates nonconscious cognition; as the number of such devices grows, so do their
19
effects on human systems. Moreover, the effects are not merely cumulative but
exponential, for increasingly devices operate not just singly but ecologically in niches
and groups.
This tectonic shift greatly magnifies the effect of the technical cognitive
nonconscious on human systems with which it interacts. The general trend is for more
and more communication to flow among intelligent devices, and relatively less among
devices and humans. In part this is because of the slow speed at which humans can
process information relative to devices, and in part because the population of devices is
growing much faster than the population of humans. The internet company Cisco
estimates that by 2015, there will 24 billion intelligent devices connected to the internet;
by contrast, the present human population of the planet is estimated at 7.1 billion.
Compared to the rate at which the human use of the internet is growing, the rate at which
intelligent devices are joining the internet is orders of magnitude higher.
As an example, consider the smart house, where the lighting system connects with
the heating system which connects with the entry/exit system and so on.
32
Because these
systems are aware of what the others are doing, they achieve a degree of coordination that
has qualitatively different effects on the human occupants than if each was separate.
Another example is the self-driving car, now in development, that has sensors and
actuators capable of monitoring the environment and reacting according. Moreover, this
capability catalyzes the development of smart roads that can communicate directly with
the car systems. Just as human cognition is massively affected by sociality, so the
nonconscious cognition of intelligent devices operates in different ways when devices
connect and communicate with one another.
2u

Interactions between Humans and the Cognitive Nonconscious of Intelligent Devices
Because computational media operate in microtemporal regimes inaccessible to
humans, some cultural critics are concerned that the missing half-second between
perception and conscious awareness may be exploited for capitalistic purposes. A grand
chess master takes about 650 milliseconds to recognize he is in checkmate; most peoples
responses, less finely tuned, require about a second or more for perceptions to register in
consciousness. By contrast, computer algorithms (in automated stock trading, for
example) can operate in the one to five millisecond range, about three orders of
magnitude faster than humans. One of the ways in which the cognitive nonconscious is
affecting human systems, then, is opening up temporal regimes in which the costs of
consciousness become more apparent and more systemically exploitable.
Luciana Parisi and Steve Goodman sketch these consequences in their discussion
of affective capitalism.
33
Affective capitalism is a parasite on the feelings,
movements, and becomings of bodies, tapping into their virtuality by investing
preemptively in futurity. Possessed by seductive brand entities you flip into autopilot, are
abducted from the present, are carried off by an array of prehensions outside
chronological time into a past not lived, a future not sensed. We term this mode of
affective programming mnemonic control, a deployment of power that exceeds current
formulations of biopower (164). In terms I have been using, computational media can
address the protoself at time scales below those at which conscious/unconscious modes of
awareness operate, so that by the time they process the protoselfs input, they are already
preconditioned to pay more attention to one consumer brand than to another. The effects
21
are similar to subliminal advertising in the 1950s and 1960s, but now, through the rapid
development of computational media, are operating at temporalities, sensory modalities,
and diverse environmental inputs that would have been unimaginable half a century ago.
Mark B. N. Hansens forthcoming book Feed Forward addresses in depth the
implications of these temporal effects of twenty-first century media.
34

Of course, not all uses of the cognitive nonconscious are exploitive or capitalistic
in their orientations. Often nonconscious cognitive devices are designed to enhance
productivity, open new avenues for research, and increase safety and well-being for
humans immersed in or affected by them, for example in the computational media
essential to the operations of major airports, where they increase safety as well as
throughput. In the case of computational media involved in the digital humanities, faster
processing speeds allow questions to be posed that simply could not have been asked or
answered using human cognition alone. As the digital humanities increasingly penetrate
the traditional humanities, misunderstandings of what computational media can and
cannot do abound, especially among scholars who have made little or no use of
computational media in their own research other than email and internet searches.
The spectrum of humanistic practices altered by the engagement with
computational media is too vast to be adequately discussed here, so I will focus on one
aspect of special interest in this journal issue: the interplay between description and
interpretation. Sharon Marcus, answering critics who contest her and coauthor Stephen
Bests call for surface reading, takes on the charge that pure description is
impossible, because every description already implicitly assumes an interpretive
viewpoint determining what details are noticed, how they are arranged and narrated, and
22
what frameworks account for them. Rather than arguing this is not the case, Marcus
turns the tables by pointing out that every interpretation necessitates description, at least
to the extent that descriptive details support, extend, and help to position the
interpretation.
35
Although not the conclusion she draws, her argument can be taken to
imply that description and interpretation are recursively embedded in one another,
description leading to interpretation, interpretation highlighting certain details over
others. Rather than being rivals of one another, then, on this view interpretation and
description are mutually supportive and entwined processes.
This helps to clarify the relation of the digital humanities to traditional modes of
understanding such as close reading and symptomatic interpretation. Many print-based
scholars see algorithmic analyses as rivals to how literary analysis has traditionally been
performed, arguing that digital-humanities algorithms are nothing more than glorified
calculating machines. But this implication misunderstands how algorithms function.
Broadly speaking, an algorithmic analysis can be either confirmatory or exploratory. For
confirmatory projects the goal is not to determine, for example, what literary drama falls
into what generic category, but rather to make explicit the factors characterizing one kind
of dramatic structure rather than another. Often new kinds of correlations appear that
raise questions about traditional criteria for genres, stimulating the search for
explanations about why these correlations pertain. When an algorithmic analysis is
exploratory, it seeks to identify patterns not previously detected by human reading, either
because the corpora is too vast to be read in its entirety, or because long-held
presuppositions constrain too narrowly the range of possibilities considered.
2S
One might suppose that algorithmic analyses are primarily descriptive rather than
interpretative, because they typically produce data about what the subject texts contain
rather than what the data mean. However, just as interpretation and description are
entwined for human readers (as Marcuss argument implies), so interpretation enters into
algorithmic analyses at several points. First, one must make some initial assumptions in
order to program the algorithms appropriately. In the case of Tom Mitchells Never-
Ending Language Learning project at Carnegie Mellon, mentioned above, the research
team first constructs ontologies to categorize words into grammatical categories. In
Timothy Lenoir and Eric Gianellas algorithms designed to detect the emergence of new
technology platforms by analyzing patent applications, they reject ontologies in favor of
determining which patent applications cite the same references.
36
The assumption here is
that co-citations will form a network of similar endeavors, and will lead to the
identification of emerging platforms. Whatever the project, the algorithms reflect initial
interpretive assumptions about what kind of data is likely to reveal interesting patterns.
Stanley Fish to the contrary, there are no all-purpose algorithms that will work in every
case.
37

Second, interpretation strongly comes into play when data are collected from the
algorithmic analysis. When Matthew Jockers found that Gothic literary texts have an
unusually high percentage of definite articles in their titles, for example, their
interpretation suggested this was so because of the prevalence of place names in the titles
(The Castle of Otranto, for example).
38
Such conclusions often lead to the choice of
algorithms for the next stage, which are interpreted in turn, and so forth in recursive
cycles.
24
Employing algorithmic analyses thus follows a similar pattern to human
description/interpretation, with the advantage that the nonconscious cognition operates
without the biases inherent in consciousness, where presuppositions can cause some
evidence to be ignored or underemphasized in favor of other evidence more in accord
with the readers own presuppositions. To take advantage of this difference, part of the
art of constructing algorithmic analyses is to keep the number of starting assumptions
small, or at least to keep them as independent as possible of the kinds of results that
might emerge. The important distinction with digital humanities projects, then, is not so
much between description versus interpretation but rather the capabilities and costs of
human reading versus the advantages and limitations of nonconscious cognition.
Working together in recursive cycles, conscious analysis and nonconscious cognition can
expand the range and significance of insights beyond what either alone can accomplish.

Staging the Cognitive Nonconscious in the Theater of Consciousness
If my hypothesis is correct about the growing importance of the cognitive
nonconscious, we should be able to detect its influence in contemporary literature and
other creative works. Of course, since these products emerge from
conscious/unconscious modes of awareness, what will be reflected is not the cognitive
nonconscious in itself, but rather its restaging within the theater of consciousness. One of
the sites where this staging is readily apparent is in contemporary conceptual poetics.
Consider, for example, Kenneth Goldsmiths uncreative writing. In Day, Goldsmith
re-typed an entire day (September 1, 2000) of the New York Times; in Fidget, he
recorded every bodily movement for a day; in Soliloquy, every word he spoke for a week
2S
(but not those spoken to him); and in Traffic, traffic reports, recorded every ten minutes
over an unnamed holiday, from a New York radio station. His work, and his
accompanying manifestos, have initiated a vigorous debate about the works value. Who,
for example, would want to read Day? Apparently not even Goldsmith himself, who
professed to type it mechanically, scarcely even looking at the page he was copying. He
often speaks of himself as mechanistic,
39
and as the most boring writer who ever
lived.
40
In his list of favored methodologies, the parallel with database technologies is
unmistakable, as he mentions information Management, word processing, databasing,
and extreme process . . . Obsessive archiving & cataloging, the debased language of
media & advertising; language more concerned with quantity than quality.
41
Of course
we might, as Marjorie Perloff does, insist there is more at work here than mere copying.
42

Still, the authors own design seems to commit him to enacting something as close to
Stanley Fishs idea of algorithmic processing as humanly possiblerote calculation,
mindless copying, mechanical repetition. It seems, in other words, that Goldsmith is
determined to stage nonconscious cognition as taking over and usurping consciousness,
perhaps simultaneously with a sly intrusion of conscious design that a reader can notice
only with some effort. That he calls the result poetry is all the more provocative, as if
the genre most associated with crafted language and the pure overflow of emotion has
suddenly turned the neural hierarchy upside down. The irony, of course, is that the
cognitive nonconscious is itself becoming more diverse, sophisticated, and cognitively
capable. Ultimately what is mimed here is not the actual cognitive nonconscious but a
parodic version that pulls two double-crosses at once, at both ends of the neuronal
spectrum: consciousness performed as if it was nonconscious, and the nonconscious
26
performed according to criteria selected by consciousness. As Perloff notes, quoting
John Cage, If something is boring after two minutes, try it for four. If still boring, try it
for eight, sixteen, thirty-two, and so on. Eventually one discovers that its not boring at
all but very interesting (157). Consciousness wearing a (distorted) mask of the
cognitive nonconscious while slyly peeping through to watch the reactionthats
interesting!
Another example of how the cognitive nonconscious is surfacing in contemporary
creative works is Kate Marshalls project on contemporary novels, which she calls
Novels by Aliens (focusing on the nonhuman as a figure, technique and desire,
Marshall shows that narrative viewpoints in a range of contemporary novels exhibit what
Fredric Jameson calls the ever-newer realisms [that] constantly have to be invented to
trace new social dynamics.
43
In Colson Whiteheads Zone One, for example, the
viewpoint for the Quiet Storms highway clearing project involves an overhead, far-away
perspective more proper to a high-flying drone than to any human observer. The
protagonist, Mark Spitz, collaborates with Quiet Storm in part because he feels lust to be
a viewpoint.
44
Although Marshall herself links these literary effects to such
philosophical movements as speculative realism, it is likely that both speculative realism
and literary experiments in nonhuman viewpoints are catalyzed by the expansive
pervasiveness of the cognitive nonconscious in the built environments of developed
countries. In this view, part of the contemporary turn toward the nonhuman is the
realization that an object need not be alive or conscious in order to function as a cognitive
agent.

Katherine Hayles 2.6.14 20.22
Deleted: new
Madigan Haley 2.6.14 15.24
Comment [1]: Please pioviue info foi this
quote, which uoesn't show up in Amazon euition
27
Reframing Interpretation
Today the humanities stand at a crossroad. On one side the path continues with
traditional understandings of interpretation, closely linked with assumptions about
humans and their relations to the world as represented in cultural artifacts. Indeed, the
majority of interpretive activities within the humanities arguably have to do specifically
with the relation of human selves to the world. This construction assumes that humans
have selves, that selves are necessary for thinking, and that selves originate in
consciousness/unconsciousness. The other path diverges from these assumptions by
enlarging the idea of cognition to include nonconscious activities. In this line of
reasoning, the cognitive nonconscious also carries on complex acts of interpretation,
which syncopate with conscious interpretations in a rich spectrum of possibilities.
What would it mean to say that the cognitive nonconscious interprets? A clue is
given by physicist Edward Fredkin, when in a seminar he casually announced, The
meaning of information is given by the processes that interpret it.
45
When Claude
Shannon first formulated information theory, Warren Weaver declared that it had nothing
to do with semantic meaning, for Shannon defined information as a function of
probability.
46
Although very significant results have followed Shannons version of
information,
47
for the humanities, a theory totally divorced from meaning has little to
contribute. Fredkins approach, however, suggests that flows of information occur within
contexts, and those contexts frequently offer multiple opportunities for interpretation.
One reason that digital technologies have become so pervasive and important is that they
are constructed to make interpretive choices as clear-cut as possible (because digital
technologies use discrete digital encoding, rather the continuous signals that analogue
28
technologies use). In many instances, however, ambiguities remain, and substantive
choices have to be made. Medical diagnostic systems, automated satellite-imagery
identification, ship navigation systems, weather-prediction programs, and a host of other
nonconscious cognitive devices interpret ambiguous information to arrive at conclusions
that rarely if ever are completely certain. Something of this kind also happens with the
protoself in humans. Integrating multiple somatic markers, the protoself too must
synthesize conflicting and/or ambiguous information to arrive at interpretations that feed
forward into the relevant brain centers, emerging as emotions, feelings, and other kinds of
awareness in core and higher consciousness, where further interpretive activities take
place.
What advantages and limitations do these two paths offer? The traditional path
carries the assumption that interpretation, requiring as it does consciousness and a self, is
confined largely if not exclusively to humans (perhaps occasionally extended to some
animals). This path reinforces the idea that humans are special, that they are the source
of almost all cognition on the planet, and that human viewpoints therefore count the most
in determining what the world means. The other path recognizes that cognition is much
broader than human thinking and that other animals as well as technical devices cognize
and interpret all the time. Moreover, it also implies that these interpretations intersect
with and very significantly influence the conscious/unconscious interpretations of
humans, which themselves depend on prior integrations and interpretations by the
protoself. The search for meaning then becomes a pervasive activity among humans,
animals, and technical devices, with many different kinds of agents contributing to a rich
ecology of collaborating, reinforcing, contesting and conflicting interpretations.
29
One of the costs of the traditional path is the isolation of the humanities from the
sciences and engineering. If interpretation is an exclusively human activity and if the
humanities are mostly about interpretation, then there are few resources within the
humanities to understand the complex embeddedness of humans in intelligent
environments and in relationships with other species. If, on the contrary, interpretation is
understood as pervasive in natural and built environments, the humanities can make
important contributions to such fields as architecture, electrical and mechanical
engineering, computer science, industrial design, and many other fields. The
sophisticated methods that the humanities have developed for analyzing different kinds of
interpretations and their ecological relationships with each other then pay rich dividends
for other fields and open onto any number of exciting collaborative projects.
Proceeding down the nontraditional path, in my view much the better choice,
requires a shift in conceptual frameworks so extensive that it might as well be called an
epistemic break. One of the first moves is to break the equivalence between thought and
cognition; another crucial move is to reconceptualize interpretation so that it applies to
information flows as well as to questions about the relations of human selves to the
world. With the resulting shifts of perspective, many of the misunderstandings about the
kinds of interventions the digital humanities are now making in the humanities simply
fade away. In closing, I want to emphasize that the issues involved here are much larger
than the digital humanities in themselves. Important as they are, focusing only on them
distorts what is at stake in talking about Interpretation and Its Rivals. The point, as far
as I am concerned, is less about methods that seem to be rivals to interpretationa
formulation that assumes interpretation and meaning are stable categories that can be
Su
adequately discussed as exclusively human activitiesthan it is about the scope and
essence of interpretation in itself.
Duke University


Notes

1
Stanis!aw Lem, Summa Technologiae, trans. Joanna Zylinska (Minneapolis: Univ. of
Minnesota Press, 2014).
2
Manual De Landa, The Geology of Morals, a Neomaterialist Interpretation, 1995,
http://www.t0.or.at/delanda/geology.htm.
3
Andy Clark, Supersizing the Mind: Embodiment, Action, and Cognitive
Extension (London: Oxford Univ. Press, 2008), 28.
4
The so-called new unconscious is understood as a mental faculty constantly surveying
the environment and communicating contextual clues that influence behavior, goals, and
priorities; see Ran Hassin, James S. Uleman, and John A. Bargh, eds., The New
Unconscious (Oxford: Oxford Univ. Press, 2005).
5
Tristram Wyatt, Pheromones and Animal Behavior: Communication by
Smell and Taste (Cambridge: Cambridge Univ. Press, 2003).
6
J. P. Gullan and P. S. Cranston, The Insects: An Outline of Entomology (New
York: Wiley-Blackwell, 2010), 321.
7
Thomas Metzinger, Being No One: The Self-Model Theory of Subjectivity
(Cambridge, MA: A Bradford Book, 2004), 107-305.
8
Owen Flanagan, Consciousness Reconsidered (Cambridge, MA: A Bradford
Book, 1993), 177.
9
Antonio Damasio, The Feeling of What Happens: Body and Emotion in the
S1

Making of Consciousness (New York: Mariner Books, 2000), 303.
10
Daniel C. Dennett, Consciousness Explained (New York: Back Bay Books, 1992),
139-70; 256-66.
11
Damasio, The Feeling of What Happens, 194.

12
Gerald M. Edelman and Giulio Tononi, A Universe of Consciousness: How Matter
Becomes Imagination (New York: Basic Books, 2000), 194.
13
You may see the video: http://www.youtube.com/watch?v=vJG698U2Mvo

14
Daniel J. Simons and Christopher S. Chabis, The Invisible Gorilla: How Our
Intuitions Deceive Us (New York: Harmony Books, 2011), 8; Simons and Chabis,
Gorillas in Our Midst, Perception 28 (1999): 1059-74.
15
Simons and Chabis, Invisible Gorilla, 59.

16
Benjamin Libet and Stephen M. Kosslyn, Mind Time: The Temporal Factor in
Consciousness (Cambridge MA: Harvard Univ. Press, 2005), 50-55.
17
Dennett, Consciousness Explained.

18
Damasio, The Feeling of What Happens, 300.

19
Damasio, The Feeling of What Happens.

20
Edelman and Tononi, A Universe of Consciousness.

21
Edelman, Neural Darwinism: The Theory of Neuronal Group Selection (New York:
Basic Books, 1987).
22
Edelman and Tononi, A Universe of Consciousness, 122-23.

23
Humberto R. Maturana and Francisco J. Varela, Autopoiesis and Cognition:
The Realization of the Living (Dordrecht: D. Reidel Publishing, 1980).
S2

24
Lawrence W. Barsalou, Grounded Cognition, Annual Review of Psychology 59
(2008): 618.
25
}ohn R. Koza, Naitin A. Keane, Natthew }. Stieetei, William Nyulowec, }essen Yu,

anu uuiuo Lanza, uenetic Piogiamming Iv: Routine Buman-Competitive Nachine

Intelligence (New Yoik anu Beilin: Spiingei 2uuS).


26
John A. Maurer, A Brief History of Algorithmic Composition,
ccrma.stanford.edu/~blackrse/algorithm.html.
27
Christopher Steiner, Automate This: How Algorithms Came to Rule Our World (New
York: Penguin, 2012), 77-83.
28
Rodney A. Brooks and Anita M. Flynn, Fast, Cheap and Out of Control: A Robot
Invasion of the Solar Systems, Journal of the British Interplanetary Society 42
(1989):478-85.
29
Matt Gardner, Partha Pratim Talukdar, Bryan Kisiel, and Tom Mitchell, Improving
Learning and Inference in a Large Knowledge-base using Latent
Syntactic Cues, Proceedings of the 2013 Conference on Empirical Methods in Natural
Language Processing (2013). http://rtw.ml.cmu.edu/rtw/publications. For NELL, see:
http://www.cmu.edu/homepage/computing/2010/fall/nell-computer-that-learns.shtml
30
Scott Patterson, Dark Pools: High-Speed Traders, A. I. Bandits, and the
Threat to the Global Financial System (New York: Crown Business, 2012).
31
Neil Johnson, Guannan Zhao, Eric Hunsader, Hong Qi, Nicholas Joshson, Jing Meng
and Brian Tivnan, Abrupt Rise in New Machine Ecology Beyond Human Response
Time, Scientific Reports: Nature Publishing Group 2627 (2013): 1-11.
SS

32
Molly Edmonds and Nathan Chandler, How Smart Homes Work, 2008.
http://home.howstuffworks.com/smart-home.htm.

33
Luciana Paiisi anu Steve uoouman, "Nuemonic Contiol," in Patiicia Ticineto
Clough anu Ciaig Willse, eus., Beyonu Biopolitics: Essays on the uoveinance of Life
anu Beath (Buiham NC: Buke 0niv. Piess, 2u11), 16S-76.


S4
Naik B. N. Bansen, !""# !%&'(&# (Chicago: 0niveisity of Chicago Piess,
foithcoming fall 2u14).

35
These remarks refer to Marcuss talk at the Interpretation and Its Rivals Conference
at the University of Virginia in September 2013.
36
Timothy Lenoir and Eric Giannell, Technology Platforms and Layers of Patent Data,
in Mario Biagioli, Peter Jaszi, and Martha Woodmansee, eds. Unmaking Intellectual
Property: Creative Production in Legal and Cultural Perspective (Chicago: Univ. of
Chicago Press, 2011), 359-84.
37
Stanley Fish, 2012. Mind Your Ps and Bs: The Digital Humanities and
Interpretation, New York Times Opinionator 2012.
http://opinionator.blogs.nytimes.com/2012/01/23/mind-your-ps-and-bs-the-digital-
humanities-and-interpretation/?_php=true&_type=blogs&_r=0.
38
Scott McLemee, Crunching Literature, Inside Higher Ed 2013.
http://www.insidehighered.com/views/2013/05/01/review-matthew-l-jockers-
macroanalysis-digital-methods-literary-history.
39
Kenneth Goldsmith, Conceptual Poetics, 2008,
http://www.poetryfoundation.org/harriet/2008/06/conceptual-poetics-kenneth-
goldsmith/?woo.
S4

40
Quoted in Marjorie Perloff, Unoriginal Genius: Poetry by Other Means in the New
Century (Chicago: Univ. of Chicago Press, 2012), 149.
41
Goldsmith, Conceptual Poetics.

42
Perloff, Unoriginal Genius, 146-65.

43
Fieuiic }ameson, "Realism anu 0topia in )*" +,&"," -&,.,/,01 S2.S-4
(SummeiFall 2u1u): SS9-S72. Please pioviue page numbei foi this quotation fiom
Antimonies of Realism.

44
Kate Marshall, The View from Above, Modern Language Convention, Jan. 14, 2014.
45
Quoted in N. Katherine Hayles, Cybernetics, in W. J. T. Mitchell and Mark B. N.
Hansen, eds., Critical Terms for Media Studies (Chicago: Univ. of Chicago Press, 2010),
145-156.
46
Claude E. Shannon and Warren Weaver, The Mathematical Theory of
Communication (New Jersey: Bell Labs, 1948).
47
James Gleick, The Information: A History, a Theory, a Flood (New York:Vintage,
2012).

Katherine Hayles 2.6.14 20.20
Formatted: Font:Italic
Katherine Hayles 2.6.14 20.21
Formatted: Font:Italic